Automatic Construction of a Chinese Sentiment Lexicon in the Field of Education Based on Word2vec
Author:Xiang Feng, Longhui Qiu, Chun Zhou,Yonghe Wu Author Unit: East China Normal University, Shanghai Engineering Research Center of Digital Educational Equipment, 3663 Zhong Shan Rd(N), Shanghai, 200062. East China Normal University, Department of education information technology, 3663 Zhong Shan Rd(N), Shanghai, 200062. East China Normal University, School of Foreign Languages, 3663 Zhong Shan Rd(N), Shanghai, 200062.
Abstract：From decision-making to intervention-implementing, sentiment analysis is applied more and more
extensively in today’s educational field. Language resources such as sentiment lexicons play a pivotal role
in analysis practices, which, unfortunately, are not readily accessible for the Chinese language. As sentiment
analysis is both domain and language specific, it is necessary to construct educational sentiment lexicons
in Chinese. By leveraging several existing generic Chinese sentiment lexicons, we aim to present a new
approach to automatically construct domain-specific Chinese sentiment lexicons for educational purposes. In
this research, we use student review corpora to train the word vectors in the first place. These word vectors
are utilized to iteratively extract semantically similar sentiment words from the general-purpose sentiment
lexicons. With this method, we managed to generate an educationally domain-specific Chinese sentiment
lexicon of 4404 words. Our experiments, based on a statistical estimation approach, demonstrate explicitly
that domain-specific lexicons outperform general-purpose lexicons when it comes to sentiment analysis in
educational scenarios.
